Human sperm detection and tracking using event-based cameras and unsupervised learning

Sperm analysis is routinely used for diagnostic and clinical purposes in the Assisted Reproductive Technology (ART). This paper presents a method for sperm detection and tracking using an event camera. The proposed approach integrates Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) for clustering sperm cells based on spatial density, K-Dimensional Tree (k-d Tree) for efficient storage and retrieval of sperm cell data, and Kalman Filtering for state estimation of each sperm cell over time. Evaluation of clustering quality is conducted using metrics such as the silhouette score and Calinski-Harabasz index, providing valuable insights into the distribution and behavior of sperm populations. Experimental results demonstrate the efficiency of the proposed framework in accurately assessing sperm head integrity when comparing with traditional clustering algorithms such as K-means Clustering, Agglomerative Clustering, and Optics Clustering algorithms. A case study was carried out by examining the trajectory and velocity of a single sperm cell. The proposed framework's ability is shown in precisely detecting and tracking sperm cells in event-based video data, with promising applications in reproductive research and clinical settings.